Repository navigation
fix(toolbox): retry op run and the toolbox action downloads - #26
Merged
Merged
Conversation
ctan run 38917887841 (2026-10-16 14:42 UTC) failed one second in: op run got a 502 from 1Password before the pipeline started. op was the one setup call with no retry, so one blip cost the hourly slot. The failure path runs inside op run, so the run also sent no /fail ping. - toolbox.yml op: up to five tries, waits 5, 10, 20, 40 s. The command first touches a marker file; once it exists, op started the command, so its exit stands and the pipeline is never started twice. - toolbox action: the go-task and op downloads use curl --retry 4 --retry-all-errors (five tries, waits 1, 2, 4, 8 s). - .github/op-retry-check.sh: stub op and sleep on PATH, run on the host. Three cases: op fails twice then starts, the command fails (no retry), op always fails (five tries). Runs in ci.yml.
This was referenced Oct 16, 2026
jshvn
added a commit
to katoptra/ctan
that referenced
this pull request
Oct 16, 2026
Moves both `uses:` pins of katoptra/lib to `e18ef7d` (v2.6.1, katoptra/lib#26). - **Toolbox action** (comes with this pin): the go-task and `op` downloads try five times, with waits of 1, 2, 4 and 8 s. - **`op` verb** (comes with the `v2` include, so it already applies after the release, with or without this PR): when `op run` fails before the pipeline starts, as with the 1Password 502 in ctan run 38917887841, it tries five times, with waits of 5, 10, 20 and 40 s. Once the pipeline has started, it never starts it again. `pipeline` is unchanged, so `render.txt` does not move. ## Test plan - [ ] `check` green on this PR (it runs the new toolbox action) - [ ] The next scheduled sync after merge is green
jshvn
added a commit
to katoptra/gnu
that referenced
this pull request
Oct 16, 2026
Moves both `uses:` pins of katoptra/lib to `e18ef7d` (v2.6.1, katoptra/lib#26). - **Toolbox action** (comes with this pin): the go-task and `op` downloads try five times, with waits of 1, 2, 4 and 8 s. - **`op` verb** (comes with the `v2` include, so it already applies after the release, with or without this PR): when `op run` fails before the pipeline starts, as with the 1Password 502 in ctan run 38917887841, it tries five times, with waits of 5, 10, 20 and 40 s. Once the pipeline has started, it never starts it again. `pipeline` is unchanged, so `render.txt` does not move. ## Test plan - [ ] `check` green on this PR (it runs the new toolbox action) - [ ] The next scheduled sync after merge is green
jshvn
added a commit
to katoptra/gnu-alpha
that referenced
this pull request
Oct 16, 2026
Moves both `uses:` pins of katoptra/lib to `e18ef7d` (v2.6.1, katoptra/lib#26). - **Toolbox action** (comes with this pin): the go-task and `op` downloads try five times, with waits of 1, 2, 4 and 8 s. - **`op` verb** (comes with the `v2` include, so it already applies after the release, with or without this PR): when `op run` fails before the pipeline starts, as with the 1Password 502 in ctan run 38917887841, it tries five times, with waits of 5, 10, 20 and 40 s. Once the pipeline has started, it never starts it again. `pipeline` is unchanged, so `render.txt` does not move. ## Test plan - [ ] `check` green on this PR (it runs the new toolbox action) - [ ] The next scheduled sync after merge is green
jshvn
added a commit
to katoptra/nongnu
that referenced
this pull request
Oct 16, 2026
Moves both `uses:` pins of katoptra/lib to `e18ef7d` (v2.6.1, katoptra/lib#26). - **Toolbox action** (comes with this pin): the go-task and `op` downloads try five times, with waits of 1, 2, 4 and 8 s. - **`op` verb** (comes with the `v2` include, so it already applies after the release, with or without this PR): when `op run` fails before the pipeline starts, as with the 1Password 502 in ctan run 38917887841, it tries five times, with waits of 5, 10, 20 and 40 s. Once the pipeline has started, it never starts it again. `pipeline` is unchanged, so `render.txt` does not move. ## Test plan - [ ] `check` green on this PR (it runs the new toolbox action) - [ ] The next scheduled sync after merge is green
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
ctan run 38917887841 (2026-10-16 14:42 UTC) failed one second into
task syncwith[ERROR] (502) Unknownfromop run --env-file=op.env. 1Password blipped, and CTAN was never contacted.op runand the toolbox action's twocurldownloads are the single-shot setup calls that every mirror makes on every run. Everything after them (rsync, aws, curl in the engine, the pings) already retries.opverb (toolbox.yml): up to five tries, with waits of 5, 10, 20 and 40 s (75 s at most). It retries only whenopfailed before the command started. The child first touches amktemp -umarker. If the marker exists,opstarted the command, its exit code stands, and the pipeline is never started twice. This coverssync,plan, the "cut off" fallback step,session-sealandempty-trash.opdownloads usecurl --retry 4 --retry-all-errors: five tries, waits of 1, 2, 4 and 8 s (curl's own doubling). Checked against a closed port: 5 attempts, 15 s..github/op-retry-check.sh: puts a stubopand a stubsleepfirst in PATH, and uses a test Taskfile that includestoolbox.ymland runsrunon the host. It needs no image and no vault. It runs in ci.yml and is listed in CLAUDE.md.oprow, the "reads the vault one time" bullet and the toolbox action paragraph mention the retry.Not covered: when
opfails all five tries, the run still sends no/failping, becauseHEALTHCHECK_URLitself comes from 1Password. The healthcheck grace still catches it.Rollout
opchange reaches every mirror whenv2moves (remote include).uses:pin moves to the new release, so each consumer gets a pin-bump PR after the tag.pipelineis unchanged, so no consumer'srender.txtmoves.Test plan
sh .github/op-retry-check.shfails on main (fail 1for the two retry cases) and passes on this branchvalidate-vars.sh --checkandchain-file.sh --checkpasstask --listparses in examples/rsync and examples/proton/bin/sh)